Phase modifier rotor fault state sensing method and system based on multi-sensor information

Abstract

The invention discloses a phase modifier rotor fault state sensing method and system based on multi-sensor information, and the method comprises the following steps: obtaining vibration signals of a plurality of sensors for detecting the vibration of a phase modifier rotor, carrying out the fusion of the vibration signals, generating a point symmetry mode scatter diagram, and inputting the point symmetry mode scatter diagram into a pre-trained convolutional neural network, according to the method, convolution operation is carried out on an input point symmetry mode scatter diagram by adopting convolution kernels of various different receptive fields through the convolutional neural network, and features obtained by convolution operation learning are fused and then classified and predicted to obtain a phase modifier rotor fault state. The convolutional neural network is trained in advance to establish a mapping relation between a point symmetry mode scatter diagram and a phase modifier rotor fault state. According to the method, data visualization, fractional-scale feature extraction and model feature learning ability enhancement are combined, the training speed and classification precision of the phase modifier rotor fault state sensing model are improved, and the model adapts to the multi-scale characteristic of a vibration signal.

Description

Technical Field

[0001] The present invention relates to the technical field of phase condenser vibration detection, and in particular to a phase condenser rotor fault state sensing method and system based on multi-sensor information. Background Art

[0002] The basic mechanical structure principle of the phase condenser rotor is similar to that of the rotor of other rotating equipment. Both of them fix the rotor on the rotating shaft and support it with bearings at both ends. However, the overall complexity of the phase condenser equipment determines that its rotor structure is relatively complex. Its characteristics are as follows: large phase condensers are large in size and weight, the operating speed exceeds the second-order critical speed, the vibration mode is complex, and it is sensitive to changes in operating conditions and operating parameters. Therefore, the phase condenser has multiple vibration states under different operating conditions. In addition, there are excitation windings on both the stator and rotor. The vibration is not a single mechanical field, but the result of the mutual coupling of electromagnetic-mechanical fields. In addition, due to its strong transient response ability and transient overload capacity, its rotor speed is high, it starts and stops suddenly, and the internal temperature field is complex and changeable, which will also affect the vibration state of the rotor. Traditional deep learning methods mainly optimize the architecture of deep networks, lack consideration of different characteristics and scales of phase condenser states, so there is a defect of insufficient accuracy in the perception of phase condenser rotor fault states. Summary of the invention

[0003] Technical problem to be solved by the present invention: In view of the above-mentioned problems in the prior art, a method and system for sensing the fault state of a phase-converter rotor based on multi-sensor information are provided. The present invention aims to combine data visualization and sub-scale feature extraction with enhanced model feature learning capabilities to improve the training speed, classification accuracy and model adaptation of the phase-converter rotor fault state perception model to the multi-scale characteristics of vibration signals.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: A method for sensing a phase condenser rotor fault state based on multi-sensor information comprises the following steps: S101, obtaining vibration signals collected by multiple sensors when a phase condenser rotor vibrates; S102, fusing multiple vibration signals to generate a point-symmetric pattern scatter plot; S103, inputting the point symmetric pattern scatter diagram into a pre-trained convolutional neural network, using the convolutional neural network to perform convolution operations on the input point symmetric pattern scatter diagram using convolution kernels with multiple different receptive fields, fusing various features obtained by the convolution operations on the convolution kernels with various different receptive fields, and classifying them to predict the fault state of the phase modulator rotor. The convolutional neural network is pre-trained to establish a mapping relationship between the point symmetric pattern scatter diagram and the fault state of the phase modulator rotor.

[0005] Optionally, in step S103, the function expression for performing the convolution operation using convolution kernels with multiple different receptive fields is: , In the above formula, Indicates l The output of the layer convolution operation, ReLU represents the linear rectifier unit activation function, N is the number of convolution kernels, Indicates l The input of the convolution operation of the layer, Indicates l Tier The feature map convolution kernels, Indicates l Layer The convolution kernels with different receptive fields refer to the convolution kernels of different sizes in the convolution operation, which result in different receptive fields.

[0006] Optionally, the convolutional neural network in step S103 is a mixed receptive field residual convolutional neural network MRFRCNN, which includes a first convolutional layer CBS, three branches, a first connection layer CONCAT, a spatial pyramid pooling fusion module SPPF and a classification module, each of the three branches includes two second convolutional layers CBS connected in series, wherein two branches also include an addition operation ADD and a third convolutional layer CBS, and the convolutional neural network uses a plurality of convolution kernels with different receptive fields to perform convolution operations on the input point symmetric pattern scatter diagram, and then fuses the features learned by the convolution operation to perform classification prediction to obtain the phase shifter rotor fault state, including: using the first convolutional layer CBS to input the point symmetric pattern scatter diagram After the point graph is convolved, it is output to three branches. The three branches perform convolution operations on the input features in sequence through two second convolutional layers CBS connected in series. The outputs of the convolution operations of two of the three branches are added ADD and output to the third convolutional layer CBS to obtain the first fusion feature. The first fusion feature is connected with the outputs of the remaining branches of the three branches through the first connection layer CONCAT to form a residual connection. The output of the first connection layer CONCAT is reduced in dimension by the spatial pyramid pooling fusion module SPPF and then classified and predicted by the classification module to obtain the fault state of the phase-shifting rotor. The convolutional layer CBS includes a convolution module, a batch normalization module and a Silu activation function module, and the convolution kernel sizes of the convolution modules in the second convolutional layer CBS of the three branches are different.

[0007] Optionally, the spatial pyramid pooling fusion module SPPF includes a fourth convolutional layer CBS, three two-dimensional maximum pooling layers, a second connection layer CONCAT and a fifth convolutional layer CBS, wherein the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are cascaded in sequence, and the outputs of the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are both used as inputs of the second connection layer CONCAT, and the output end of the second connection layer CONCAT is connected to the classification module through the fifth convolutional layer CBS.

[0008] Optionally, the classification module includes two fully connected layers and a Softmax layer connected in sequence, and the output of the spatial pyramid pooling fusion module SPPF passes through the two fully connected layers and the Softmax layer in sequence to obtain a classification result.

[0009] Optionally, before step S103, the step of training a convolutional neural network is also included: S201, obtaining vibration signals of a phase condenser rotor under different phase condenser rotor fault states detected by multiple sensors, and fusing the multiple vibration signals to generate a point symmetric mode scatter plot sample; S202, adding labels of phase shifter rotor fault states to point symmetric pattern scatter plot samples to obtain a training sample data set, and then dividing the training sample data set into a training set and a test set; S203, using the training set to train the convolutional neural network, and in each round of training, the training sample data in the training set is input into the convolutional neural network and the classification prediction loss Loss is calculated by a preset loss function, and the classification prediction loss Loss is used to update the network parameters of the convolutional neural network, until the number of training rounds is equal to the preset threshold N, and then jumping to step S204; S204, use the network parameters of the trained convolutional neural network to test the test set. If the test accuracy meets the requirements, it is determined that the convolutional neural network training is completed; otherwise, jump to step S201 to continue training the convolutional neural network.

[0010] Optionally, in step S102, fusing the multiple vibration signals to generate a point-symmetric pattern scatter diagram includes mapping the multiple vibration signals in the form of scattered points in a polar coordinate diagram, and the function expression of the polar coordinates of the mapped scattered points is: , , , In the above formula, and are the angles of deflection of the point in the counterclockwise and clockwise directions with respect to the mirror symmetry plane, respectively; is the polar diameter of the point; is the deflection angle of the mirror symmetry plane, ,in ; is the number of symmetry planes, and are the maximum amplitude and minimum amplitude in the original time domain waveform of the vibration signal, is the gain coefficient and satisfies < .

[0011] In addition, the present invention also provides a phase regulator rotor fault state perception system based on multi-sensor information, comprising: A data acquisition program unit, used for acquiring vibration signals collected by multiple sensors when the phase modulator rotor vibrates; A data fusion program unit, used for fusing multiple vibration signals to generate a point-symmetric pattern scatter plot; A classification prediction program unit is used to input a point symmetric pattern scatter diagram into a pre-trained convolutional neural network, and the convolutional neural network uses a plurality of convolution kernels with different receptive fields to perform convolution operations on the input point symmetric pattern scatter diagram, and fuses various features obtained by the convolution operations on the convolution kernels with different receptive fields and classifies them to predict the fault state of the phase modulator rotor. The convolutional neural network is pre-trained to establish a mapping relationship between the point symmetric pattern scatter diagram and the fault state of the phase modulator rotor.

[0012] Optionally, in step S103, the function expression for performing the convolution operation using convolution kernels with multiple different receptive fields is: , In the above formula, Indicates l The output of the layer convolution operation, ReLU represents the linear rectifier unit activation function, N is the number of convolution kernels, Indicates l The input of the convolution operation of the layer, Indicates l Tier The feature map convolution kernels, Indicates l Layer The convolution kernels with different receptive fields refer to the convolution kernels of different sizes in the convolution operation, which result in different receptive fields.

[0013] Optionally, the convolutional neural network of the classification prediction program unit is a mixed receptive field residual convolutional neural network MRFRCNN, which includes a first convolutional layer CBS, three branches, a first connection layer CONCAT, a spatial pyramid pooling fusion module SPPF and a classification module, each of the three branches includes two second convolutional layers CBS connected in series, wherein two branches also include an addition operation ADD and a third convolutional layer CBS, the convolutional neural network uses a plurality of convolution kernels with different receptive fields to perform convolution operations on the input point symmetric pattern scatter diagram, and then fuses the features learned by the convolution operation to perform classification prediction to obtain the phase shifter rotor fault state, including: using the first convolutional layer CBS to input the point symmetric pattern scatter diagram The scatter plot is convolved and then output to three branches. The three branches perform convolution operations on the input features in sequence through two second convolutional layers CBS connected in series. The outputs of the convolution operations of two of the three branches are added ADD and then output to the third convolutional layer CBS to obtain a first fusion feature. The first fusion feature is connected with the outputs of the remaining branches of the three branches through the first connection layer CONCAT to form a residual connection. The output of the first connection layer CONCAT is reduced in dimension by a spatial pyramid pooling fusion module SPPF and then classified and predicted by a classification module to obtain the fault state of the phase-shifting rotor. The convolutional layer CBS includes a convolution module, a batch normalization module and a Silu activation function module, and the convolution kernel sizes of the convolution modules in the second convolutional layers CBS of the three branches are different.

[0014] Optionally, the spatial pyramid pooling fusion module SPPF includes a fourth convolutional layer CBS, three two-dimensional maximum pooling layers, a second connection layer CONCAT and a fifth convolutional layer CBS, wherein the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are cascaded in sequence, and the outputs of the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are both used as inputs of the second connection layer CONCAT, and the output end of the second connection layer CONCAT is connected to the classification module through the fifth convolutional layer CBS.

[0015] Optionally, the classification module includes two fully connected layers and a Softmax layer connected in sequence, and the output of the spatial pyramid pooling fusion module SPPF passes through the two fully connected layers and the Softmax layer in sequence to obtain a classification result.

[0016] Optionally, the following program modules for training convolutional neural networks are also included: A sample acquisition program module is used to acquire vibration signals of a phase condenser rotor detected by multiple sensors under different phase condenser rotor fault states, and fuse the multiple vibration signals to generate a point symmetric mode scatter plot sample; A sample construction program module is used to add labels of phase-shifting rotor fault states to point-symmetric pattern scatter plot samples to obtain a training sample data set, and then divide the training sample data set into a training set and a test set; A network training program module is used to train the convolutional neural network using a training set, and in each round of training, the training sample data in the training set is input into the convolutional neural network and a preset loss function is used to calculate the classification prediction loss Loss, and the classification prediction loss Loss is used to update the network parameters of the convolutional neural network, until the number of training rounds is equal to a preset threshold value N, and then the step S204 is jumped; The network testing program module is used to test the test set using the network parameters of the trained convolutional neural network. If the test accuracy meets the requirements, the convolutional neural network training is determined to be complete; otherwise, the sample construction program module is jumped to continue training the convolutional neural network.

[0017] Optionally, the data fusion program unit fuses multiple vibration signals to generate a point-symmetric pattern scatter diagram, including mapping the multiple vibration signals in the form of scattered points in a polar coordinate diagram, and the function expression of the polar coordinates of the mapped scattered points is: , , , In the above formula, and are the angles of deflection of the point in the counterclockwise and clockwise directions with respect to the mirror symmetry plane, respectively; is the polar diameter of the point; is the deflection angle of the mirror symmetry plane, ,in ; is the number of symmetry planes, and are the maximum amplitude and minimum amplitude in the original time domain waveform of the vibration signal, is the gain coefficient and satisfies < .

[0018] In addition, the present invention also provides a phase modulator rotor fault state perception system based on multi-sensor information, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the phase modulator rotor fault state perception method based on multi-sensor information.

[0019] In addition, the present invention also provides a computer-readable storage medium, in which a computer program / instruction is stored, and the computer program / instruction is programmed or configured to execute the phase-shifting rotor fault state perception method based on multi-sensor information through a processor.

[0020] In addition, the present invention also provides a computer program product, including a computer program / instruction, which is programmed or configured to execute the phase-converter rotor fault state perception method based on multi-sensor information through a processor.

[0021] Compared with the prior art, the present invention mainly has the following advantages: the method of the present invention comprises the following steps: obtaining vibration signals of a phase condenser rotor detected by multiple sensors and fusing them to generate a point symmetric pattern scatter diagram, and then inputting the signals into a pre-trained convolutional neural network; the convolutional neural network uses a plurality of convolution kernels with different receptive fields to perform convolution operations on the input point symmetric pattern scatter diagram, and then fuses the features learned by the convolution operation to perform classification prediction to obtain the phase condenser rotor fault state; the convolutional neural network is pre-trained to establish a mapping relationship between the point symmetric pattern scatter diagram and the phase condenser rotor fault state; the present invention proposes a phase condenser rotor fault state perception method based on multi-sensor information fusion point symmetric pattern scatter diagram combined with a multi-scale convolutional neural network, combines data visualization and sub-scale feature extraction with model feature learning ability enhancement, improves the training speed, classification accuracy and model adaptation of the phase condenser rotor fault state perception model of the phase condenser rotor to the multi-scale characteristics of the vibration signal, improves the rotor fault training speed, classification accuracy and the angle of the model adaptation to the multi-scale characteristics of the vibration signal, can accurately identify the phase condenser rotor fault, and improves the reliability of the phase condenser. BRIEF DESCRIPTION OF THE DRAWINGS In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 Schematic diagram of the basic flow of the method of the embodiment of the present invention.

[0023] Figure 2 Schematic diagram of the perception principle flow of the convolutional neural network in an embodiment of the present invention.

[0024] Figure 3 It is a principle diagram of the point-symmetric pattern scatter diagram used in the embodiment of the present invention.

[0025] Figure 4 This is an architecture diagram of the mixed receptive field residual convolutional neural network MRFRCNN in an embodiment of the present invention.

[0026] Figure 5 4 is a structural diagram of the spatial pyramid pooling fusion module SPPF in an embodiment of the present invention.

[0027] Figure 6 This is a diagram of the recognition accuracy of the mixed receptive field residual convolutional neural network MRFRCNN training set and test set in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Among them, the drawings are only used for illustrative explanations and represent only schematic diagrams, and should not be understood as limitations of the present invention; in order to better illustrate the embodiments of the present invention, the drawings may be omitted and do not represent the sizes of actual products; for those skilled in the art, it is understandable that some descriptions in the drawings may be omitted.

[0030] like Figure 1 As shown, the phase modulator rotor fault state perception method based on multi-sensor information in this embodiment includes the following steps: S101, obtaining vibration signals collected by multiple sensors when a phase condenser rotor vibrates; S102, fusing multiple vibration signals to generate a point-symmetric pattern scatter plot; S103, inputting the point symmetric pattern scatter diagram into a pre-trained convolutional neural network, using the convolutional neural network to perform convolution operations on the input point symmetric pattern scatter diagram using convolution kernels with multiple different receptive fields, fusing various features obtained by the convolution operations on the convolution kernels with various different receptive fields, and classifying them to predict the fault state of the phase modulator rotor. The convolutional neural network is pre-trained to establish a mapping relationship between the point symmetric pattern scatter diagram and the fault state of the phase modulator rotor.

[0031] The specific method of detecting the vibration signal of the phase condenser rotor when vibrating by multiple sensors in step S101 of this embodiment is to use sensors (such as vibration sensors) set at different positions on the surface of the phase condenser outlet end bearing and the non-outlet section bearing, and collect the vibration signal of the phase condenser rotor when vibrating by using the sensors. Specifically, the method of this embodiment uses multiple sets of eddy current sensors to detect the vibration signal of the phase condenser body shaft when it vibrates relative to each other, and the sensors at the phase condenser bearings are installed at 90 degrees to each other, and multiple sets of velocity vibration sensors are used to detect the absolute vibration of the phase condenser body bearings, and the sensors at the phase condenser bearings are installed at 90 degrees to each other.

[0032] When multiple vibration signals are fused to generate a point-symmetric pattern scatter plot in step S102 of this embodiment, sample data is first extracted using a resampling method based on the collected original vibration signal of the camera; then the symmetric dot pattern (SDP) algorithm is used to perform overall feature fusion on the vibration information detected by multiple sensors of the camera, so as to more comprehensively map out the features of the identified object. The symmetric dot pattern SDP algorithm can clearly map complex time series in the form of scattered points in a polar coordinate graph, which can improve the visualization ability of the original time domain signal in a graphical way. Due to the particularity of polar coordinate images, multi-source information can be fused in a limited area through the SDP method. In polar coordinates and In the time domain, they are Moment and The amplitude at the moment, the original signal can be converted into polar coordinate space through the SDP method The points in Figure 2 As shown, in step S102, fusing multiple vibration signals to generate a point-symmetric pattern scatter diagram includes mapping the multiple vibration signals in the form of scattered points in a polar coordinate diagram, and the function expression of the polar coordinates of the mapped scattered points is: , , , In the above formula, and are the angles of deflection of the point in the counterclockwise and clockwise directions with respect to the mirror symmetry plane, respectively; is the polar diameter of the point; is the deflection angle of the mirror symmetry plane, ,in ; is the number of symmetry planes, and are the maximum amplitude and minimum amplitude in the original time domain waveform of the vibration signal, is the gain coefficient and satisfies < After the one-dimensional signal is transformed by the symmetrical point pattern SDP algorithm, a "petal-shaped" scatter plot (point symmetrical pattern scatter plot) can be obtained, which can avoid data compression and feature loss caused by traditional signal processing methods. The differences between different types of signals are mainly manifested in: 1) the curvature of the scatter petals; 2) the distribution and shape characteristics of the scatter points; 3) the geometric center position. In the same polar coordinate system, information fusion of multiple signals can be achieved in different quadrants, which further highlights the differences.

[0033] In step S103 of this embodiment, the function expression of using convolution kernels with multiple different receptive fields to perform convolution operation is: , In the above formula, Indicates l The output of the layer convolution operation, ReLU represents the linear rectifier unit activation function, N is the number of convolution kernels, Indicates l The input of the convolution operation of the layer, Indicates l Tier The feature map convolution kernels, Indicates l Layer Convolution kernels with different biases and different receptive fields refer to the different sizes of convolution kernels in the convolution operation, resulting in different receptive fields. Different convolution kernels capture different image information, and usually, the larger the convolution kernel, the larger the receptive field, the more overall information is seen, and the better the global features are captured, but it will lead to a surge in the amount of calculation. Conversely, the smaller the convolution kernel, the smaller the receptive field, and the local information obtained will be more specific.

[0034] Based on the above-mentioned convolution operation using convolution kernels with a variety of different receptive fields, the required convolutional neural network structure can be adopted.

[0035] The vibration coupling of the various components of the phase shifter will cause the rotor vibration signal to show multi-scale characteristics, and the SDP image contains multi-sensor synchronization information. In order to obtain more comprehensive features and improve the recognition accuracy of the convolutional neural network model, this embodiment introduces a residual module into the multi-receptive field convolutional neural network, and uses different convolution kernels to jointly extract the phase shifter vibration features. The convolutional neural network in step S103 of this embodiment is a mixed receptive field residual convolutional neural network MRFRCNN, such as Figure 3As shown, the mixed receptive field residual convolutional neural network MRFRCNN includes a first convolutional layer CBS, three branches, a first connection layer CONCAT, a spatial pyramid pooling fusion module SPPF and a classification module. Each of the three branches includes two second convolutional layers CBS connected in series, wherein two branches also include an addition operation ADD and a third convolutional layer CBS. The convolutional neural network uses convolution kernels with multiple different receptive fields to perform convolution operations on the point-symmetric pattern scatter diagram of the input (Input), and then fuses the features learned by the convolution operation to perform classification prediction to obtain the fault state of the phase-shifting rotor, including: performing a convolution operation on the input point-symmetric pattern scatter diagram through the first convolutional layer CBS and outputting it to the three branches. , the three branches perform convolution operations on the input features in sequence through two second convolutional layers CBS connected in series, the outputs of the convolution operations of two of the three branches are added ADD and then output to the third convolutional layer CBS to obtain the first fusion feature, the first fusion feature is connected with the outputs of the remaining branches of the three branches through the first connection layer CONCAT to form a residual connection, the output of the first connection layer CONCAT is reduced in dimension by the spatial pyramid pooling fusion module SPPF and then classified and predicted by the classification module to obtain the fault state of the phase regulator rotor; the convolutional layer CBS includes a convolution module, a batch normalization module and a Silu activation function module, and the convolution kernel sizes of the convolution modules in the second convolutional layer CBS of the three branches are different. For example, as an optional implementation, in this embodiment, they are 1×1, 3×3 and 5×5 respectively, Figure 3 The five numbers after the convolution layer CBS represent the number of channels, the number of kernels, the size of the convolution kernel (two digital elements) and the step size. Assuming that the input layer is a 128*128 original image, after passing through the first 3×3 convolution kernel CBS layer, the data is divided into three branches, namely: branch 1: 1*1 convolution CBS branch; branch 2: 3×3 CBS branch; branch 3: 5×5 CBS branch. Because the receptive fields of each branch are different, vibration features can be extracted from multiple scales to achieve the effect of feature fusion. The three branches all use two CBS layers, with 16 and 32 kernels and a step size of 2. The outputs of branch 2 and branch 3 are added, and then the data flow is stabilized by passing through the 1*1 point convolution CBS layer, and then the CONCAT operation is performed with the output of branch 1, that is, a residual connection is formed with branch 1, which reduces the complexity of the model to prevent overfitting, and at the same time prevents gradient disappearance or gradient explosion. After the three branches are CONCATed, they are reduced in dimension through the spatial pyramid pooling fusion module SPPF module, which uses multiple windows to extract spatial feature information of different sizes, thus improving the robustness of the model to spatial layout and object variability. Figure 3As shown in the figure, the classification module of this embodiment includes two fully connected layers (FClayer 1 and FC layer 2) and a Softmax layer connected in sequence. The output of the spatial pyramid pooling fusion module SPPF passes through two fully connected layers (FC layer 1 and FC layer 2) and a Softmax layer in sequence to obtain a classification result. The numbers after the fully connected layers FClayer 1 and FC layer 2 are the number of channels. Figure 3 The Output in represents the final classification result.

[0036] The mixed receptive field residual convolutional neural network MRFRCNN of this embodiment changes the operation of adding activation function after the traditional convolution kernel to the convolution layer CBS, including: convolution, Bach Normalization, Silu activation function, that is, Figure 3 In CBS, data is standardized and weights are effectively activated after different convolution features are extracted; on this basis, the mixed receptive field residual convolutional neural network MRFRCNN of this embodiment also uses the spatial pyramid pooling fusion module SPPF to replace the traditional pooling layer, which can effectively avoid image distortion caused by cropping and scaling operations of image regions from a global perspective, and improve the classification accuracy. Figure 4 As shown in the figure, the spatial pyramid pooling fusion module SPPF includes a fourth convolutional layer CBS, three two-dimensional maximum pooling layers (MaxPool 2d), a second connection layer CONCAT and a fifth convolutional layer CBS, wherein the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are cascaded in sequence, and the outputs of the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are used as inputs of the second connection layer CONCAT, and the output end of the second connection layer CONCAT is connected to the classification module through the fifth convolutional layer CBS.

[0037] See also Figure 5 , before step S103, this embodiment also includes the step of training the convolutional neural network: S201, obtaining vibration signals of a phase condenser rotor under different phase condenser rotor fault states detected by multiple sensors, and fusing the multiple vibration signals to generate a point symmetric mode scatter plot sample; S202, adding labels of phase shifter rotor fault states to point symmetric pattern scatter plot samples to obtain a training sample data set, and then dividing the training sample data set into a training set and a test set; S203, using the training set to train the convolutional neural network, and in each round of training, the training sample data in the training set is input into the convolutional neural network and the classification prediction loss Loss is calculated by a preset loss function, and the classification prediction loss Loss is used to update the network parameters of the convolutional neural network, until the number of training rounds is equal to the preset threshold N, and then jumping to step S204; S204, use the network parameters of the trained convolutional neural network to test the test set. If the test accuracy meets the requirements, it is determined that the convolutional neural network training is completed; otherwise, jump to step S201 to continue training the convolutional neural network.

[0038] The samples used for testing are input into the trained model for fault diagnosis. The working state and fault type of the phase condenser rotor can be determined based on the output of the mixed receptive field residual convolutional neural network MRFRCNN model. The recognition accuracy (using accuracy as an indicator) obtained by the training set and the test set in different iteration rounds of this embodiment is as follows: Figure 6 As shown, see Figure 6 It can be seen that the accuracy of the method in this embodiment is close to 100% at the 47th iteration round. It can be seen that the model of the method in this embodiment can converge quickly on the one hand, and can accurately identify the fault state of the phase-converter rotor on the other hand.

[0039] In order to verify the phase condenser rotor fault state perception method based on multi-sensor information in this embodiment, the experimental simulation phase condenser rotor operating states in this embodiment are divided into: normal, rotor rubbing, and rotor imbalance. 200 sample data for each of the three states are set and randomly divided into training sets and test sets, as shown in Table 1 below.

[0040] Table 1 Composition of training set and test set

[0041] In the PyTorch environment, a multi-receptive field convolutional neural network was built for diagnostic classification, and the network learning rate was set to 0.0001, the number of batch samples was 32, and the iteration was 50 rounds. Figure 5 As shown in the figure, after the model training is completed, the recognition accuracy of the test set reaches 98.67%. From the figure, we can see the effect curve of the model on the training set and the test set. The model reaches a high accuracy of 96% around the 13th round and begins to converge. The entire training process is stable, and there is no overfitting or underfitting. It can be seen that the phase-converter rotor fault state perception method based on multi-sensor information in this embodiment can effectively realize the rotor fault state perception by using the constructed mixed receptive field residual convolutional neural network with high stability.

[0042] The method of this embodiment (SDP-MRFCNN) is compared with the existing STFT-CNN and SDP-CNN methods. The STFT-CNN method uses short-time Fourier transform to convert the signal into a time-frequency diagram and inputs it into the traditional convolutional neural network CNN for diagnosis. The SDP-CNN method inputs the point-symmetric pattern scatter diagram of multi-sensor fusion information into the traditional convolutional neural network CNN structure for diagnosis and classification. The results are shown in Table 2.

[0043] Table 2 Comparative experimental results

[0044] It can be seen from Table 2 that, compared with the existing STFT-CNN and SDP-CNN methods, the recognition accuracy of the method in this embodiment (SDP-MRFCNN) is the highest, reaching 98.67%, indicating that the improvement (MRFCNN) made by the method in this embodiment (SDP-MRFCNN) on the traditional convolutional neural network CNN is an effective improvement.

[0045] To summarize, in order to improve the accuracy of phase shifter rotor fault identification, the phase shifter rotor fault state perception method based on multi-sensor information in this embodiment includes collecting phase shifter rotor vibration signals, using symmetrical point patterns to perform overall feature fusion on the vibration information of multiple sensors of the phase shifter, constructing a mixed receptive field residual convolutional neural network MRFRCNN, and using multi-scale convolution kernels with different receptive fields to perform global feature learning and diagnosis on the point symmetrical pattern graph. A phase shifter rotor fault state perception method based on multi-sensor information fusion point symmetrical pattern scatter graph and multi-scale convolutional neural network is proposed. By introducing a residual module into the multi-receptive field convolutional neural network, different convolution kernels are used to jointly extract phase shifter vibration features, so as to obtain more comprehensive features and improve the recognition accuracy of the convolutional neural network model, thereby improving the reliability of the phase shifter.

[0046] In addition, this embodiment also provides a phase modulator rotor fault state perception system based on multi-sensor information, including: A data acquisition program unit, used for acquiring vibration signals collected by multiple sensors when the phase modulator rotor vibrates; A data fusion program unit, used for fusing multiple vibration signals to generate a point-symmetric pattern scatter plot; A classification prediction program unit is used to input a point symmetric pattern scatter diagram into a pre-trained convolutional neural network, and the convolutional neural network uses a plurality of convolution kernels with different receptive fields to perform convolution operations on the input point symmetric pattern scatter diagram, and fuses various features obtained by the convolution operations on the convolution kernels with different receptive fields and classifies them to predict the fault state of the phase modulator rotor. The convolutional neural network is pre-trained to establish a mapping relationship between the point symmetric pattern scatter diagram and the fault state of the phase modulator rotor.

[0047] In this embodiment, the function expression of the convolution operation performed using convolution kernels with multiple different receptive fields in step S103 is: , In the above formula, Indicates l The output of the layer convolution operation, ReLU represents the linear rectifier unit activation function, N is the number of convolution kernels, Indicatesl The input of the convolution operation of the layer, Indicates l Tier The feature map convolution kernels, Indicates l Layer The convolution kernels with different receptive fields refer to the convolution kernels of different sizes in the convolution operation, which result in different receptive fields.

[0048] In this embodiment, the convolutional neural network of the classification prediction program unit is a mixed receptive field residual convolutional neural network MRFRCNN, which includes a first convolutional layer CBS, three branches, a first connection layer CONCAT, a spatial pyramid pooling fusion module SPPF and a classification module. Each of the three branches includes two second convolutional layers CBS connected in series, wherein two branches also include an addition operation ADD and a third convolutional layer CBS. The convolutional neural network uses convolution kernels with multiple different receptive fields to perform convolution operations on the input point symmetric pattern scatter diagram, and then fuses the features learned from the convolution operation to perform classification prediction to obtain the phase shifter rotor fault state, including: using the first convolutional layer CBS to input the point symmetric pattern scatter diagram The scatter plot is convolved and then output to three branches. The three branches perform convolution operations on the input features in sequence through two second convolutional layers CBS connected in series. The outputs of the convolution operations of two of the three branches are added ADD and then output to the third convolutional layer CBS to obtain a first fusion feature. The first fusion feature is connected with the outputs of the remaining branches of the three branches through the first connection layer CONCAT to form a residual connection. The output of the first connection layer CONCAT is reduced in dimension by a spatial pyramid pooling fusion module SPPF and then classified and predicted by a classification module to obtain the fault state of the phase-shifting rotor. The convolutional layer CBS includes a convolution module, a batch normalization module and a Silu activation function module, and the convolution kernel sizes of the convolution modules in the second convolutional layers CBS of the three branches are different.

[0049] In this embodiment, the spatial pyramid pooling fusion module SPPF includes a fourth convolutional layer CBS, three two-dimensional maximum pooling layers (MaxPool 2d), a second connection layer CONCAT and a fifth convolutional layer CBS, wherein the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are cascaded in sequence, and the outputs of the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are all used as inputs of the second connection layer CONCAT, and the output end of the second connection layer CONCAT is connected to the classification module through the fifth convolutional layer CBS.

[0050] In this embodiment, the classification module includes two fully connected layers and a Softmax layer connected in sequence.

[0051] This embodiment also includes the following program modules for training convolutional neural networks: A sample acquisition program module is used to acquire vibration signals of a phase condenser rotor detected by multiple sensors under different phase condenser rotor fault states, and fuse the multiple vibration signals to generate a point symmetric mode scatter plot sample; A sample construction program module is used to add labels of phase-shifting rotor fault states to point-symmetric pattern scatter plot samples to obtain a training sample data set, and then divide the training sample data set into a training set and a test set; A network training program module is used to train the convolutional neural network using a training set, and in each round of training, the training sample data in the training set is input into the convolutional neural network and a preset loss function is used to calculate the classification prediction loss Loss, and the classification prediction loss Loss is used to update the network parameters of the convolutional neural network, until the number of training rounds is equal to a preset threshold value N, and then the step S204 is jumped; The network testing program module is used to test the test set using the network parameters of the trained convolutional neural network. If the test accuracy meets the requirements, the convolutional neural network training is determined to be complete; otherwise, the sample construction program module is jumped to continue training the convolutional neural network.

[0052] In this embodiment, the data fusion program unit fuses multiple vibration signals to generate a point-symmetric pattern scatter diagram, including mapping the multiple vibration signals in the form of scattered points in a polar coordinate diagram, and the function expression of the polar coordinates of the mapped scattered points is: , , , In the above formula, and are the angles of deflection of the point in the counterclockwise and clockwise directions with respect to the mirror symmetry plane, respectively; is the polar diameter of the point; is the deflection angle of the mirror symmetry plane, ,in ; is the number of symmetry planes, and are the maximum amplitude and minimum amplitude in the original time domain waveform of the vibration signal, is the gain coefficient and satisfies < .

[0053] In addition, this embodiment also provides a phase-modulator rotor fault state perception system based on multi-sensor information, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the phase-modulator rotor fault state perception method based on multi-sensor information.

[0054] In addition, this embodiment also provides a computer-readable storage medium, in which a computer program / instruction is stored. The computer program / instruction is programmed or configured to execute the phase-shifting rotor fault state perception method based on multi-sensor information through a processor.

[0055] In addition, this embodiment also provides a computer program product, including a computer program / instruction, which is programmed or configured to execute the phase-converter rotor fault state perception method based on multi-sensor information through a processor.

[0056] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0057] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for sensing a phase condenser rotor fault state based on multi-sensor information, characterized in that: The steps include: S101, obtaining vibration signals collected by multiple sensors when a phase condenser rotor vibrates; S102, fusing multiple vibration signals to generate a point-symmetric pattern scatter plot; S103, inputting the point symmetric pattern scatter diagram into a pre-trained convolutional neural network, using the convolutional neural network to perform convolution operations on the input point symmetric pattern scatter diagram using convolution kernels with multiple different receptive fields, fusing various features obtained by the convolution operations on the convolution kernels with various different receptive fields, and classifying them to predict the fault state of the phase modulator rotor. The convolutional neural network is pre-trained to establish a mapping relationship between the point symmetric pattern scatter diagram and the fault state of the phase modulator rotor.

2. The method for sensing phase condenser rotor fault status based on multi-sensor information according to claim 1 is characterized in that: The function expression of the convolution operation performed using convolution kernels with multiple different receptive fields in step S103 is: ， In the above formula, Indicates l The output of the layer convolution operation, ReLU represents the linear rectifier unit activation function, N is the number of convolution kernels, Indicates l The input of the convolution operation of the layer, Indicates l Tier The feature map convolution kernels, Indicates l Layer The convolution kernels with different receptive fields refer to the convolution kernels of different sizes in the convolution operation, which result in different receptive fields.

3. The method for sensing phase condenser rotor fault status based on multi-sensor information according to claim 2 is characterized in that: The convolutional neural network in step S103 is a mixed receptive field residual convolutional neural network MRFRCNN, which includes a first convolutional layer CBS, three branches, a first connection layer CONCAT, a spatial pyramid pooling fusion module SPPF and a classification module. Each of the three branches includes two second convolutional layers CBS connected in series, wherein two branches also include an addition operation ADD and a third convolutional layer CBS. The convolutional neural network uses convolution kernels with multiple different receptive fields to perform convolution operations on the input point symmetric pattern scatter diagram, and then fuses the features learned by the convolution operation to perform classification prediction to obtain the phase shifter rotor fault state, including: using the first convolutional layer CBS to input the point symmetric pattern scatter diagram After the convolution operation, the output is output to three branches. The three branches perform convolution operations on the input features in sequence through two second convolution layers CBS connected in series. The outputs of the convolution operations of two of the three branches are added ADD and then output to the third convolution layer CBS to obtain the first fusion feature. The first fusion feature is connected with the output of the remaining branch of the three branches through the first connection layer CONCAT to form a residual connection. The output of the first connection layer CONCAT is reduced in dimension by the spatial pyramid pooling fusion module SPPF and then classified and predicted by the classification module to obtain the fault state of the phase-shifting rotor. The convolution layer CBS includes a convolution module, a batch normalization module and a Silu activation function module, and the convolution kernel sizes of the convolution modules in the second convolution layer CBS of the three branches are different.

4. The method for sensing phase condenser rotor fault status based on multi-sensor information according to claim 3 is characterized in that: The spatial pyramid pooling fusion module SPPF includes a fourth convolutional layer CBS, three two-dimensional maximum pooling layers, a second connection layer CONCAT and a fifth convolutional layer CBS, wherein the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are cascaded in sequence, and the outputs of the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are all used as inputs of the second connection layer CONCAT, and the output end of the second connection layer CONCAT is connected to the classification module through the fifth convolutional layer CBS.

5. The method for sensing phase condenser rotor fault status based on multi-sensor information according to claim 3 is characterized in that: The classification module includes two fully connected layers and a Softmax layer connected in sequence, and the output of the spatial pyramid pooling fusion module SPPF passes through the two fully connected layers and the Softmax layer in sequence to obtain a classification result.

6. The method for sensing phase condenser rotor fault status based on multi-sensor information according to claim 1, characterized in that: Before step S103, the step of training the convolutional neural network is also included: S201, obtaining vibration signals of a phase condenser rotor detected by multiple sensors under different phase condenser rotor fault states, and fusing the multiple vibration signals to generate a point symmetric mode scatter plot sample; S202, adding labels of phase condenser rotor fault states to point symmetric pattern scatter plot samples to obtain a training sample data set, and dividing the training sample data set into a training set and a test set; S203, using the training set to train the convolutional neural network, and in each round of training, the training sample data in the training set is input into the convolutional neural network and the classification prediction loss Loss is calculated using a preset loss function, and the classification prediction loss Loss is used to update the network parameters of the convolutional neural network, until the number of training rounds is equal to the preset threshold N, then jumping to step S204; S204, use the network parameters of the trained convolutional neural network to test the test set. If the test accuracy meets the requirements, it is determined that the convolutional neural network training is completed; otherwise, jump to step S201 to continue training the convolutional neural network.

7. The method for sensing phase condenser rotor fault status based on multi-sensor information according to claim 1, characterized in that: In step S102, fusing multiple vibration signals to generate a point-symmetric pattern scatter diagram includes mapping the multiple vibration signals in the form of scattered points in a polar coordinate diagram, and the function expression of the polar coordinates of the mapped scattered points is: ， ， ， In the above formula, and are the angles of deflection of the point in the counterclockwise and clockwise directions with respect to the mirror symmetry plane, respectively; is the polar diameter of the point; is the deflection angle of the mirror symmetry plane, ,in ; is the number of symmetry planes, and are the maximum amplitude and minimum amplitude in the original time domain waveform of the vibration signal, is the gain coefficient and satisfies < .

8. A phase condenser rotor fault state perception system based on multi-sensor information, characterized in that: include: A data acquisition program unit, used for acquiring vibration signals collected by multiple sensors when the phase modulator rotor vibrates; A data fusion program unit, used for fusing multiple vibration signals to generate a point-symmetric pattern scatter plot; A classification prediction program unit is used to input a point symmetric pattern scatter diagram into a pre-trained convolutional neural network, and the convolutional neural network uses a plurality of convolution kernels with different receptive fields to perform convolution operations on the input point symmetric pattern scatter diagram, and fuses various features obtained by the convolution operations on the convolution kernels with different receptive fields and classifies them to predict the fault state of the phase modulator rotor. The convolutional neural network is pre-trained to establish a mapping relationship between the point symmetric pattern scatter diagram and the fault state of the phase modulator rotor.

9. The phase modulator rotor fault state perception system based on multi-sensor information according to claim 8 is characterized in that: The classification prediction program unit uses a variety of convolution kernels with different receptive fields to perform convolution operations. The function expression is: ， In the above formula, Indicates l The output of the layer convolution operation, ReLU represents the linear rectifier unit activation function, N is the number of convolution kernels, Indicates l The input of the convolution operation of the layer, Indicates l Tier The feature map convolution kernels, Indicates l Layer The convolution kernels with different receptive fields refer to the convolution kernels of different sizes in the convolution operation, which result in different receptive fields.

10. The phase modulator rotor fault state perception system based on multi-sensor information according to claim 9 is characterized in that: The convolutional neural network of the classification prediction program unit is a mixed receptive field residual convolutional neural network MRFRCNN, which includes a first convolutional layer CBS, three branches, a first connection layer CONCAT, a spatial pyramid pooling fusion module SPPF and a classification module. Each of the three branches includes two second convolutional layers CBS connected in series, wherein two branches also include an addition operation ADD and a third convolutional layer CBS. The convolutional neural network uses a plurality of convolution kernels with different receptive fields to perform convolution operations on the input point symmetric pattern scatter diagram, and then fuses the features learned from the convolution operation to perform classification prediction to obtain the fault state of the phase shifter rotor, including: using the first convolutional layer CBS to perform convolution operations on the input point symmetric pattern scatter diagram; After the point graph is convolved, it is output to three branches. The three branches perform convolution operations on the input features in sequence through two second convolutional layers CBS connected in series. The outputs of the convolution operations of two of the three branches are added ADD and output to the third convolutional layer CBS to obtain the first fusion feature. The first fusion feature is connected with the outputs of the remaining branches of the three branches through the first connection layer CONCAT to form a residual connection. The output of the first connection layer CONCAT is reduced in dimension by the spatial pyramid pooling fusion module SPPF and then classified and predicted by the classification module to obtain the fault state of the phase-shifting rotor. The convolutional layer CBS includes a convolution module, a batch normalization module and a Silu activation function module, and the convolution kernel sizes of the convolution modules in the second convolutional layer CBS of the three branches are different.

11. The phase modulator rotor fault state perception system based on multi-sensor information according to claim 10, characterized in that: The spatial pyramid pooling fusion module SPPF includes a fourth convolutional layer CBS, three two-dimensional maximum pooling layers, a second connection layer CONCAT and a fifth convolutional layer CBS, wherein the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are cascaded in sequence, and the outputs of the fourth convolutional layer CBS and the three two-dimensional maximum pooling layers are all used as inputs of the second connection layer CONCAT, and the output end of the second connection layer CONCAT is connected to the classification module through the fifth convolutional layer CBS.

12. The phase modulator rotor fault state perception system based on multi-sensor information according to claim 10, characterized in that: The classification module includes two fully connected layers and a Softmax layer connected in sequence, and the output of the spatial pyramid pooling fusion module SPPF passes through the two fully connected layers and the Softmax layer in sequence to obtain a classification result.

13. The phase modulator rotor fault state perception system based on multi-sensor information according to claim 8, characterized in that: It also includes the following program modules for training convolutional neural networks: A sample acquisition program module is used to acquire vibration signals of a phase condenser rotor detected by multiple sensors under different phase condenser rotor fault states, and fuse the multiple vibration signals to generate a point symmetric mode scatter plot sample; A sample construction program module is used to add labels of phase-shifting rotor fault states to point-symmetric pattern scatter plot samples to obtain a training sample data set, and then divide the training sample data set into a training set and a test set; A network training program module is used to train the convolutional neural network using a training set, and in each round of training, the training sample data in the training set is input into the convolutional neural network and a preset loss function is used to calculate the classification prediction loss Loss, and the classification prediction loss Loss is used to update the network parameters of the convolutional neural network, until the number of training rounds is equal to a preset threshold value N, and then the step S204 is jumped; The network testing program module is used to test the test set using the network parameters of the trained convolutional neural network. If the test accuracy meets the requirements, it is determined that the convolutional neural network training is completed; Otherwise, jump to the sample construction program module to continue training the convolutional neural network.

14. The phase modulator rotor fault state perception system based on multi-sensor information according to claim 8, characterized in that: The data fusion program unit fuses multiple vibration signals to generate a point-symmetric pattern scatter diagram, including mapping the multiple vibration signals in the form of scattered points in a polar coordinate diagram, and the function expression of the polar coordinates of the mapped scattered points is: ， ， ， In the above formula, and are the angles of deflection of the point in the counterclockwise and clockwise directions with respect to the mirror symmetry plane, respectively; is the polar diameter of the point; is the deflection angle of the mirror symmetry plane, ,in ; is the number of symmetry planes, and are the maximum amplitude and minimum amplitude in the original time domain waveform of the vibration signal, is the gain coefficient and satisfies < .

15. A phase condenser rotor fault state perception system based on multi-sensor information, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the phase modulator rotor fault state perception method based on multi-sensor information as described in any one of claims 1 to 7.

16. A computer-readable storage medium having a computer program / instruction stored therein, characterized in that: The computer program / instruction is programmed or configured to execute the method for sensing the fault state of a phase-converter rotor based on multi-sensor information as described in any one of claims 1 to 7 through a processor.

17. A computer program product comprising a computer program / instructions, characterized in that The computer program / instruction is programmed or configured to execute the method for sensing the fault state of a phase-converter rotor based on multi-sensor information as described in any one of claims 1 to 7 through a processor.

Images